Publication Details
Abstract
The rapid spoilage of perishable foods such as fish and tomato poses significant challenges to food safety and contributes to food waste. This study evaluated the effectiveness of artificial intelligence (AI) models in predicting spoilage and supporting timely interventions. Fresh fish and tomato samples (eight each) were collected from local markets and farms, stored under refrigerated (4°C) and ambient (25°C) conditions, while environmental parameters—temperature, humidity, and light exposure were continuously monitored using digital sensors. Microbial counts were determined at regular intervals using standard microbiological techniques, and intrinsic food properties (pH, moisture content, water activity) were measured. The AI models—Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Network (ANN) were trained on normalized datasets integrating food and environmental variables to predict microbial load (CFU/g) and spoilage onset. Model performance was evaluated using accuracy, precision, recall, F1-score, and RMSE, and real-time predictions were compared with actual measurements. Results indicated that microbial growth accelerated at 25°C, reaching severe spoilage within 24–48 hours, while refrigerated samples exhibited slower growth. Among the AI models, Random Forest consistently achieved the highest accuracy (95% for fish, 93% for tomato), precision, recall, F1-score, and the lowest RMSE, accurately predicting spoilage with errors below 3%. ANN also performed well in capturing temporal patterns, whereas SVM showed moderate predictive capability. In conclusion, AI models, particularly Random Forest, effectively forecasted microbial growth and spoilage, enabling early interventions to improve food safety and reduce waste. Recommendations include maintaining low storage temperatures, implementing AI-based monitoring, integrating real-time sensor data for dynamic prediction, and promoting AI adoption in perishable food supply chains.